linux-kernel - Re: [PATCH v1 3/5] random: add helpers for random numbers with given floor or range

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 15 Nov 2022 09:42:25 +0100
From:   Yann Droneaud <ydroneaud@...eya.com>
To:     "Jason A. Donenfeld" <Jason@...c4.com>
Cc:     Kees Cook <keescook@...omium.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Herbert Xu <herbert@...dor.apana.org.au>,
        Theodore Ts'o <tytso@....edu>, linux-crypto@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1 3/5] random: add helpers for random numbers with given
 floor or range

Hi,

Le 14/11/2022 à 19:38, Jason A. Donenfeld a écrit :
> On Mon, Nov 14, 2022 at 07:04:13PM +0100, Yann Droneaud wrote:
>> I have a bad feeling about this one, and can't help but thinking it's going
>> to bite someone: when asked to pick a number *between* 0 and 10,
>> I usually think I'm allowed to pick 10 (even if I'm going to answer 7 as it should).
> This is one of those bikeshed things you see all over the place, like
> whether slices in a language should be [start index, end index] or
> [start index, length], or whether arrays should be 0-based or 1-based.
> We'll never settle this variety of dispute here.
>
> But in this case, there are some particular reasons why it must be this
> way. Firstly, usage of it this way matches most of the ways the function
> is actually used in the kernel, and fits existing semantics. This alone
> I find compelling. But also, having all of these functions use half-open
> intervals means that each function can take care of its entire range,
> without having to resort to using 64-bit arithmetic, and no function is
> a complete subset of any other function. So doing it this way makes
> these maximally useful too.

For get_random_below(), which replaces a modulo, <bikeshedding> and could
have been called get_random_mod()</bikeshedding>, having an open upper
range seems fine. It's already what can be achieved by the % operator.

But I believe it's unfortunate get_random_between() cannot be called to
get a number up to UINT32_MAX, as get_random_between(0, UINT32_MAX) would
be capped to UINT32_MAX - 1.

When not a constant, one could hope the function can cope with a maximum
that would grow up to and including UINT32_MAX.


> So anyway I think the function has to be defined like this. If you'd
> like to bikeshed over a different name than "between", though, be my
> guest. Maybe you'd like "from" better. But probably "between" is fine,
> and with enough good examples (as my conversion patch does) and the
> clear succinct documentation comment, we should be good.

That the conversion patch [1] that triggered my comment: I find replacing
the following rather unpleasing, somewhat uncanny:

-get_random_u32_below(1024) + 1 + get_random_u32_between(1, 1024 + 1) I 
would prefer
- get_random_u32_below(1024) + 1 + get_random_u32_between(1, 1024) [1]https://lore.kernel.org/lkml/20221114164558.1180362-4-Jason@zx2c4.com/

Regards.

-- 
Yann Droneaud
OPTEYA